The proposed research will examine the ability of infants, children, and adults to learn non-adjacent regularities in langauge. Almost all aspects of language (phonology, morphology, syntax, and semantics) contain structural relations between elements that are reliable and predictable and yet obstructed by other elements. Common examples include the relationship between "is" and "-ing" from morphology, or the relationship between noun phrases and verb phases in syntax. To date, theories of language learning have not adequately addressed how learners acquire structural knowledge of this type. The difficulty of learning these types of relationsips is one of the biggest criticisms of statistical learning theories of language acquisition, which largely involve learning language by tracking transitional probabilities between adjacent elements in language. The work is also especially important because failure to learn these types of non-adjacent regularities is a hallmark of many language disorders like Specific Language Impariment (SLI). The first aim of this research is to show how language learners might acquire knowledge about non- adjacent dependancies using knowledge of the language's hiearchical structure, and to see if this structure is learnable by an enriched form of statistical learning. We hypothesize that learners will be able to discover probability-dependant "chunks" in language, and use these chunks to transform non-adjacent relationships into adjacent relationships. This aim will be addressed by behavioral experiments with infants and adults. The second aim is to see if people find it easier to learn non-adjacent dependancies in language when those dependancies can be tied to language-external semantic cues. For example, a verb (or a verb phrase) and a direct object (or its noun phrase) might be tightly coupled linguisticly, even if they are separated by embedded clauses in particular sentences. It may be that the way learners acquire this linguistic regularity is by keeping track of the semantic or conceptual association between the verb's and direct object's real world referents. We hypothesize that in conditions where there is semantic relatedness between nonadacent elements in language, this will faciliate learning of the linguistic non-adjacent regularity. This aim will be addressed by a behavioral experiment with children and by a computational model designed to explore how the interaciton of semantic and linguistic knowledge facilliates learning. In summary, proposed research will address a critical question about how people learn the structure of language, address a key shortcoming of one current theory of language acquistion, and have important implications for language disorders like SLI.